aigon AGENTS.md

Aigon — Agent Instructions

Aigon is a spec-driven multi-agent harness. Keep this always-loaded file concise; read the linked references only when the task touches their area.

Start Here

• aigon-cli.js is dispatch only. Command handlers live in lib/commands/; shared behavior lives in focused lib/ modules.
• Read the active feature/research spec before implementation and follow its scope and validation instructions.
• New to this codebase? Read docs/code-tour.md — annotated verbatim excerpts of the core logic in execution order.
• Read docs/architecture.md when changing modules, workflow state, installation, agent sessions, or the dashboard.
• Read docs/testing.md before choosing validation commands or adding tests.
• Read .aigon/docs/development_workflow.md for the feature/research lifecycle and agent-status protocol.
• Read docs/seeds.md before touching a seed repo. The two-repo seed architecture makes an ordinary-looking reset capable of wiping work.

Public Repository

This repository is public and everything in it ships to users. Never commit a literal credential, API key, or private roster — use placeholders or environment variables. .githooks/pre-commit blocks staged .env and *.local files; it is not a secret scanner, so the check is yours to make.

Target Repositories: Zero Opinion

Aigon owns only its feature/research/feedback process, specs, .aigon/ state, worktrees, branches, and managed sessions. It must not assume a target repo's language, package manager, test framework, build/deploy process, directory layout, or commit conventions.

This is load-bearing for everything under templates/{generic,docs,specs,prompts,skill-pointers}/, which is installed into unknown repositories. If template wording would be wrong in a Python monorepo, Rust crate, Go service, or static site, generalise it. Per-worktree setup is operator-declared through worktreeSetup; Aigon does not detect a stack or inject install commands.